mRNA destabilization is the dominant effect of mammalian microRNAs by the time substantial repression ensues

Negligible contribution of nuclear or deadenylated RNA to TE changes

The adaptation of ribosome profiling to mammalian cells has provided a sensitive and quantitative method to assess the influence of miRNAs on TE (Guo et al., 2010). Ribosome profiling uses high-throughput sequencing of ribosome-protected fragments (RPFs) to determine the positions of millions of ribosomes on mRNAs (Ingolia et al., 2009). To assess the TE of a gene, RPFs mapping to its open reading frame are normalized to its mRNA abundance, as determined by RNA-seq.

When comparing samples with and without a particular miRNA, the change in RPFs for a target of that miRNA reflects the aggregate effects of mRNA degradation and translational repression, while the change in mRNA reflects only the component attributable to degradation. After accounting for the change in RPFs attributed to mRNA degradation, the residual change in RPFs reflects a change in TE, which is interpreted as the miRNA-mediated translational repression acting on the message at the moment the ribosomes were arrested.

Previously, we observed little miRNA-mediated translational repression in mammalian cells, with the concern that these modest TE changes might actually be overestimates (Guo et al., 2010). An overestimation would occur if some polyadenylated mRNA were sequestered away from the compartment containing both miRNAs and ribosomes, as would be the case for mRNAs awaiting export from the nucleus. In this case, miRNA-mediated degradation of mRNAs only in the cytoplasm would lead to a larger relative loss of RPFs (which are only from the cytoplasm) than mRNA fragments (which are from both the nucleus and cytoplasm), thereby inflating the apparent translational repression. To address this concern, we performed ribosome profiling on miRNA- and mock-transfected U2OS cells, and in parallel performed RNA-seq on poly(A)-selected RNA from both whole-cell lysates and cytoplasmic fractions. The efficacy of fractionation was demonstrated by the depletion of pre-ribosomal RNAs (pre-rRNAs) in the cytoplasmic fraction (Figure S1A). Following transfection of miR-1, a miRNA not normally expressed in U2OS cells, repression was observed, with significant degradation of mRNAs with at least one miR-1 3′-UTR site (Figure 1A). The amount of degradation was indistinguishable in the RNA-seq libraries made with either whole-cell or cytoplasmic mRNA, and thus the amount of translational repression was similarly indistinguishable (Figure 1A). The same was observed with miR-155, another miRNA not normally expressed in U2OS cells, demonstrating that a nuclear mRNA-sequestration artifact does not detectably elevate the signal for miRNA-mediated translational repression in mammalian cells.

A second concern involved the measurement of poly(A)-selected RNA. Monitoring changes in poly(A)-selected RNA leaves unanswered the question of whether repressed mRNAs are degraded or merely deadenylated, and under-recovery of partially deadenylated messages during poly(A)-selection might overestimate the amount of mRNA degradation that has occurred. To address this concern, we generated a third set of RNA-seq libraries from the aforementioned U2OS cells, starting with whole-cell RNA preparations that were not poly(A)-selected and instead were depleted of both tRNAs and rRNAs. Greatly increased RNA-seq coverage of replication-dependent histone mRNAs, which lack poly(A) tails, illustrated our ability to detect RNAs regardless of poly(A)-tail length (Figure S1B). Results for miRNA-dependent changes in tRNA/rRNA-depleted RNA were similar to those of poly(A)-selected RNA (Figure 1A), which indicated that changes in accumulation of mRNA refractory to poly(A) selection were negligible. These results imply that the absolute amount of deadenylated mRNAs and other intermediates underrepresented in poly(A)-selected RNA is small, even for repressed mRNAs, presumably because these decay intermediates are rapidly decapped and degraded. Thus, concerns that translational repression measurements might have been either under- or over-estimates appear to be unfounded; comparing TEs calculated by simply normalizing RPF changes to those of poly(A)-selected RNA accurately measures translational repression in mammalian cells.

To aid comparisons, the results in Figure 1A can be summarized in compound bar graphs (Figure 1B). For each experiment, the mean RPF fold change (distance that the compound bar extends below zero) indicates the overall repression. The mRNA contribution (blue component of the compound bar) indicates the extent to which mRNA degradation explains this repression, and any residual RPF change is the TE contribution (green component), which reflects the translational repression of the remaining mRNA. Based on the RPF reductions attributable to these two repression modes, their relative contributions to repression are then calculated (Figure S2). Of the two modes, mRNA degradation dominates in U2OS cells (Figure 1B), despite the presence of a P-body subtype reported to impart increased translational repression (Castilla-Llorente et al., 2012).

Dominant mRNA destabilization in many contexts

We expanded our analysis to examine the steady-state effects of gaining or losing a miRNA in additional cell lines and biological contexts. These experiments included studies comparing RPF and mRNA measurements in liver from wild-type mice, which expresses miR-122, to those in liver from mice lacking the mir-122 gene. Similarly, the effects of miR-155 in activated primary murine B cells were measured comparing cells from wild-type mice to those lacking the mir-155 gene. These loss-of-function experiments enabled analysis of endogenous targets in their endogenous settings. The effects on predicted targets of endogenous miR-122 in mouse liver, endogenous miR-155 in primary mouse B cells, induced miR-1 (expressed from a transgene) in 3T3 cells, and transfected miR-1 in HEK293T cells all resembled the published effects of endogenous miR-223 in neutrophils and transfected miRNAs in HeLa cells (Figure 1C). In all settings, reduced mRNA levels explained most of the steady-state RPF reduction observed in the presence of the miRNA, implying that miRNAs predominantly act to reduce target mRNA levels. Nonetheless, mean RPF reduction attributable to translational repression was observed, ranging from 1–34% of the total, depending on the experiment.

Because a 7–8-nt site to a miRNA is not always sufficient to mediate miRNA targeting, high-throughput proteomic measurements can be used to identify high-confidence targets by identifying site-containing genes with less protein in the presence of the miRNA (Guo et al., 2010). With this in mind, we performed a quantitative proteomics experiment using SILAC (stable isotope labeling with amino acids in culture) to identify a set of genes with reduced protein after inducing miR-1 in 3T3 cells (Table S1) and pulsed SILAC (Selbach et al., 2008) to identify those responding to miR-155 in activated B cells (Table S2). These proteomics-supported predicted targets showed greater mean repression than did the complete set of genes with ≥1 site, as expected if they were enriched in direct targets of the miRNA (Figure 2C). For new and published experiments with proteomics-supported predicted targets, the fractional repression attributed to translational repression ranged from 6–26%, somewhat narrower than the range observed when considering all mRNAs with sites, perhaps because a focus on the more confidently identified targets decreased experimental variability.

Although the amount of repression attributed to translational repression did not always reach statistical significance, our results are consistent with the idea that a small amount of translational repression occurs for each direct target in each context. As was found previously (Baek et al., 2008; Hendrickson et al., 2009; Guo et al., 2010), a gene-by-gene analysis of results from each of the newly examined settings revealed no compelling evidence for a subset of genes repressed at only the translational level (Figure 1D), although the possibility of a few such genes cannot be ruled out.

Matching mRNA and proteomic results for less proficient miRNAs

In pilot experiments aimed at extending our studies to other endogenous contexts, we used wild-type and miRNA-deleted mice to acquire mRNA microarray data for macrophages and neutrophils with and without miR-21, and B cells with and without miR-150. Although these miRNAs were each among the most frequently sequenced miRNAs in the respective wild-type cells (Figure S3A), we observed weak miRNA effects when comparing sets of genes with and without 3′-UTR sites to the cognate miRNA (Figure S3B).

A potential explanation for the weak signals observed by mRNA profiling was that most of the repression was occurring through translational repression rather than mRNA degradation. However, when we used quantitative proteomics to test this possibility, the proteomics results mirrored those of the mRNA arrays, providing no evidence for substantial translational repression (Figure S3B and Table S3). Thus the weak repression signals observed at the mRNA level for endogenously expressed miR-21 and miR-150 were not due to a discrepancy between mRNA changes and the overall effects of miRNA-mediated repression. These results add to the growing list of endogenous settings for which mRNA changes accurately represent the effects of miRNA-mediated repression. This list now includes miR-223 in neutrophils (Baek et al., 2008; Guo et al., 2010), miR-21 in macrophages and neutrophils (Figure S3B), miR-122 in liver (Figure 1C), miR-150 in primary B cells (Figure S3B), and miR-155 in activated B cells (Figure 1C).

Dynamics of endogenous mRNA repression by inducible miRNAs

The shifting regulatory regime in the early zebrafish embryo, which changes the consequences of miRNA-mediated poly(A)-tail shortening, confounded the previous attempt to determine the dynamics of the two modes of repression for endogenous messages (Bazzini et al., 2012; Subtelny et al., 2014). Therefore, we set out to characterize the regulatory dynamics of miRNA-mediated repression of endogenous mRNAs and determine if there might be an endogenous setting in which these dynamics could give rise to a phase of substantial translation-dominated repression, as previously observed in reporter experiments (Bethune et al., 2012; Djuranovic et al., 2012).

Perhaps the most dynamic mammalian miRNA is miR-155, which is rapidly and strongly induced in B and T cells upon activation (Thai et al., 2007). In primary murine B cells, we observed a nearly 10-fold increase 4 h after activation with lipopolysaccharide, IL-4, and anti-CD40 (Figure 2A). Although presumably not as strong as for miR-430 in zebrafish embryos [which is expressed from as many as 93 loci (Giraldez et al., 2005)], miR-155 induction was nonetheless stronger than that of other mammalian miRNAs, in that no other mammalian miRNA has been reported to increase so rapidly to a high level of expression.

To assess the dynamics of translational repression and mRNA decay during miR-155–mediated repression, we isolated B cells from wild-type and miR-155 knockout mice, activated these cells, and then performed ribosome profiling and RNA-seq to monitor miRNA-dependent TE and mRNA changes occurring soon after induction. At 2 h post-activation, repression of genes with ≥1 site was detectable, but neither the mRNA nor the TE component was significantly decreased on its own. At 4 h, the small amount of repression was predominantly attributable to reduced TE (Figure 2B). By 8 h, the proportion attributed to translational repression abated, and at this time point the mRNA degradation so closely approached overall repression that the mean mRNA change for genes with ≥1 site slightly exceeded the mean RPF change (Figure 2B; P = 0.028, two-tailed K–S test for TE). Because this slight excess was not observed for the proteomics-supported predicted targets or in similar experiments with other miRNAs, we attribute it to experimental variability rather than evidence of translational activation. After 48 h, mRNA degradation continued to dominate (Figure 2B; as already shown in the steady-state analyses of Figure 1C), which indicated that B cells resemble other cells with respect to steady-state repression.

Although we found some evidence for translational repression dominating early in miR-155 induction, the amount of repression observed during this brief period was much weaker than that observed during the analogous phase of reporter experiments. Thus, we cannot claim to have found a mammalian setting with an early phase of substantial translational repression of endogenous messages, i.e., a time at which substantial repression would be missed if only mRNA changes were monitored. To further explore repression dynamics in mammalian cells, we created stable, miRNA-inducible 3T3 cell lines in which doxycycline treatment rapidly induced the expression of a miRNA not normally expressed in 3T3 cells (either miR-1, miR-124, or miR-155) to levels comparable to those of miR-21 and the let-7 miRNA family (Figure 2A), which are the miRNA and miRNA family most frequently sequenced for these cells (Rissland et al., 2011). The major advantage of such cell lines for studying the dynamics of translational repression and decay on endogenous messages is that, in contrast to B cells, miRNA induction does not accompany significant developmental changes, allowing the miRNA effects to be more easily isolated. With these lines we performed ribosome profiling and RNA-seq soon after miRNA induction, comparing translational efficiencies and mRNA expression levels with those of un-induced cells.

To account for the 2–3 h lag prior to the appearance of increased mature miRNA, the first time-point examined was 4 h post-induction. At 4 h, the miR-155–expressing line showed significant repression of genes with ≥1 site, all of which was attributed to translational repression (Figure 2C). At later time points, mRNA degradation dominated, as observed in B cells. For the miR-1–expressing line, 4 h was too early to observe significant repression for genes with ≥1 site, and by 8 h, mRNA degradation already dominated (Figure 2C), suggesting that we had missed any potential translation-dominant phase. For miR-124, a translation-dominant phase also was not observed (Figure 2C), presumably because induction was too gradual to achieve significant repression at early time points. (As we did not acquire murine proteomics data for miR-124, the top predicted targets were used instead of proteomics-supported targets.) Because miRNA induction in vivo is rarely more rapid than that achieved for miR-124 in our inducible line, we suggest that the miR-124 results are representative of most endogenous settings.

Minimal influence of translational stress and state

Having investigated eight different cell types and six different miRNAs, and having considered both pre-steady-state and later time points without identifying a setting with substantial overall repression in which translational effects dominated, we turned to the potential influence of cellular state. Studies of lin-4–mediated repression in C. elegans suggest that starvation might tip the balance towards more translational inhibition with less mRNA degradation (Holtz and Pasquinelli, 2009), presumably because starvation influences global translational activity. Therefore, we compared the relative contributions of TE and mRNA degradation for 3T3 cells in three translational states: 1) dividing cells, which have very active translation [polysome to monosome ratio (P:M) = 11.6], 2) contact-inhibited cells (P:M = 1.4), and 3) contact-inhibited cells under Torin1-induced mTOR inhibition (P:M = 0.4) (Figure 3A). We found no pervasive difference in the relative contribution of translational repression to miR-1– and miR-155–mediated repression between these states (Figure 3B), despite the ~30-fold range in translational activity. Thus, translational stress, and more generally the translational state, does not have a perceptible global impact on the mode of miRNA-mediated regulation in these mammalian cells.

Because translating ribosomes displace miRNA-directed silencing complexes, which renders miRNA sites in the path of the ribosome much less effective than those ≥15 nt downstream of the stop codon (Grimson et al., 2007), we reasoned that the efficacy of sites within open reading frames (ORFs) might increase in conditions of reduced translational activity. Indeed, relative to the efficacy of 3′-UTR sites, the efficacy of ORF sites did appear to increase when translation was repressed with Torin1 (Figure S4A), which supported the model in which displacement of bound miRNAs by translating ribosomes is the predominant reason that ORF sites are ineffective.

The principles of repression dynamics

Our results in 3T3 and B cells, considered in light of the fundamental differences between the nature of translational repression and mRNA destabilization, lead to the following principles regarding the miRNA-mediated repression of endogenous mRNAs in mammalian cells: Compared to translational repression, detectable mRNA destabilization occurs after more of a lag, presumably because mRNA decay takes longer than inhibiting translation initiation. Because of this relative lag, after unusually robust miRNA induction, we can detect a short phase resembling that observed in reporter experiments, in which most of the repression is from decreased TE. However, the lag in destabilization does not last long, and destabilization soon dominates. To illustrate these principles, we simulated the repression time course of a rapidly induced miRNA for which 80% of the steady-state repression is through mRNA destabilization and 20% is through translational repression (Figure 4). In our simulation, translational repression begins immediately upon miRNA–mRNA association, and mRNA degradation occurs through an increased degradation rate for the miRNA-bound mRNA. This approach yields an early phase in which translational repression dominates, consistent with that observed in our experimental time courses (Figure 4B). The transition from mostly translational repression to mostly mRNA destabilization takes place at 5.7 h (Figure 4C), when relatively little overall repression (9.7% RPF decrease, compared to a 50% decrease at steady state) is occurring (Figure 4B). Our example simulates very rapid miRNA induction; within 6 h the induced miRNA reaches levels that would make it the highest expressed miRNA in 3T3 cells (Figure 4A), similar to or faster than the induction observed in our 3T3 cell lines (Figure 2A). Slowing the induction rate by about half would result in this transition occurring at a point of even less repression (6.6% RPF decrease), and thus in most mammalian contexts, miRNA induction would be too slow to yield detectable repression during the phase in which TE changes dominate. For an early phase of substantial repression mediated primarily through TE changes, miRNA induction would have to be stronger than that ever reported, which is consistent with our inability to find a mammalian context with substantial translation-based repression.

Decreases in mRNA and TE lead to decreased protein from the targeted messages, and this change in protein is what matters to the cell. Despite the ultimate importance of the protein changes, measuring these changes over time is less informative for analyzing miRNA repression dynamics than is measuring RPF and mRNA changes, which more directly captures the molecular effects of the miRNA in inhibiting translation and destabilizing mRNA. RPF and mRNA measurements are also more suitable for quantitative comparisons for two reasons: 1) they enabled accurate comparisons of more miRNA targets, and 2) they were each acquired using analogous methods that measured differences at one moment in time without the complications that arise from pre-steady-state measurements of protein changes. With regard to these complications, protein differences detected using direct labeling or standard metabolic labeling (e.g., SILAC) cannot distinguish between protein synthesized before or after induction of the miRNA and thus are unsuitable for pre-steady-state measurements because they would underestimate the impact on newly synthesized protein. Pulsed SILAC differentiates between pre-existing and newly synthesized protein but as currently implemented still entails an extended period (≥6 h for global measurements) of metabolic labeling (Schwanhausser et al., 2009; Huo et al., 2012), which compromises its utility for observing the results of the first few hours of repression.

Despite the advantages of measuring RPF and mRNA changes, we note that during pre-steady-state conditions, the relative TE and mRNA effects can underestimate the relative contribution of translational repression to miRNA-mediated repression at the protein level. For example, at a given time point reduced mRNA might explain 80% of the RPF effect, leaving only 20% of the reduced protein synthesis at that moment to be explained by translational repression, but when considering the reduced protein levels (not current protein synthesis) more repression might have been due to translational repression. This is because the reduced protein levels are a function of the miRNA effects integrated up to the current time point, which includes earlier periods in which translational repression might have represented a greater share of the decreased protein synthesis.

The extent to which the relative contribution of translational repression would be underestimated depends on three factors: 1) the extent to which translational repression represents a greater share of the overall repression at the earlier time periods, 2) the relative strength of the overall repression during earlier periods, and 3) the stability of the protein. Our results indicate that with respect to the second factor, the relative strength of the overall repression during earlier periods is low in mammalian contexts, which implies that any underestimate of the contribution of translational repression to the reduction in protein levels would be minimal. In our simulation the greatest underestimate was observed at 5.7 h, when TE changes explained 49% of the reduction in protein synthesis at that moment and 58% of the reduction in protein accumulation, assuming intermediate protein stability (10 h protein half-life; Figure 4C). A shorter protein half-life further diminished the small differential between protein synthesis and protein accumulation (Figure 4C), whereas a longer half-life delayed the onset of any consequential miRNA effect on protein abundance to a period well beyond the onset of substantial mRNA decay (Figure 4B). In sum, monitoring protein levels rather than TE would not increase the prospects for finding a mammalian setting in which substantial translational repression dominates.

Comparison of fish embryos and mammalian contexts

Attempts to characterize the dynamics of the two modes of miRNA-mediated repression in zebrafish embryos were confounded by two unique features of fish and frog embryos prior to gastrulation: 1) a strong coupling between poly(A)-tail length and translational efficiency and 2) an unusual mRNA metabolism wherein mRNAs with short poly(A)-tails are stable. These features do not necessarily preclude analysis of dynamics, but in these contexts changes in TE due to miRNA-mediated deadenylation must be accounted for independently of changes in TE due to direct miRNA-mediated translational repression. Indeed, when the repression due to mRNA decay is thought of as including deadenylation-dependent translational repression, mRNA decay is the predominant mode of miRNA-mediated repression at all time points analyzed in zebrafish (Subtelny et al., 2014) just as it is at all but the earliest time points in mammalian cells. An important difference between most mammalian systems and early developmental systems (and presumably neuronal synapses or other systems with the aforementioned features) is that, in the latter, effects on translation must be measured to accurately capture the impact of the miRNA on gene expression, and effects on deadenylation must be measured to understand how repression is achieved. However, neither system seems to have a phase in which deadenylation-independent translational repression performs substantial repression without even stronger repression detectable by mRNA changes.

Mechanistic interpretations

Although translational repression and mRNA decay both lead to reduced protein synthesis, the mechanism used for repression has important biological implications. To the extent that repression occurs through translational repression, rapid recovery would be possible without requiring new transcription. This would, for example, be the case in early zebrafish embryos, where the repression of miRNA targets could be rapidly reversed through cytoplasmic polyadenylation. In most settings, however, reversal of miRNA-mediated repression requires new transcription, as mRNA decay constitutes the major mode of repression.

When miRNA-mediated mRNA decay was first reported, it was proposed to occur either through active recruitment of mRNA-degradation machinery or as a secondary effect of inhibiting translation (Lim et al., 2005). Although we observe translational repression prior to the decay of endogenous mRNAs in some experiments, this temporal relationship does not imply that mRNA decay is a consequence of translational repression, because it is also consistent with mRNA decay simply being a slower process. Indeed, several observations favor the model that the decay occurs through active recruitment of mRNA-degradation machinery rather than as a secondary effect of inhibiting translation. First, miRNA targeting can destabilize reporter transcripts that cannot be translated, which indicates that mRNA destabilization is not merely a secondary effect of reducing the number of ribosomes translating an mRNA (Mishima et al., 2006; Wu et al., 2006; Eulalio et al., 2007; Wakiyama et al., 2007; Eulalio et al., 2009; Fabian et al., 2009), although it does not rule out models in which only translationally repressed mRNAs can be destabilized. Second, direct biochemical interactions link miRNAs to Argonaute, Argonaute to TNRC6, and TNRC6 to the deadenylase complexes (the PAN2-PAN3 complex and the CCR4-NOT complex) that shorten the poly(A) tail (Braun et al., 2012), thereby showing how the mRNA-degradation machinery can be actively recruited independent of either the act or the consequence of translational repression. Finally, our work greatly expands the number of mammalian systems examined and shows that in each of these systems mRNA destabilization explains a large majority (from 66% to >90%) of the miRNA-mediated repression observed at steady state.

The idea that the mRNA destabilization might be a secondary consequence of inhibiting translation would be more plausible if a larger fraction of the steady-state repression was through translational repression; otherwise, the mRNA destabilization is out of proportion to the translational repression. We are not aware of any mammalian examples in which translationally repressed messages are so destabilized as a secondary consequence of this repression that the amount of steady-state destabilization exceeds the amount of steady-state translational repression. Indeed, the idea that mammalian messages might be destabilized solely as a secondary consequence of reduced ribosome occupancy or density appears to be largely an extrapolation from observations made in bacteria and yeast, but not mammalian cells (Muhlrad et al., 1995; Schwartz and Parker, 1999; Deana and Belasco, 2005). When examining mammalian mRNAs in general (irrespective of miRNA targeting), we find only a very weak correlation between TE and mRNA half-life (Figure S4B, R² = 0.004 and 0.001 for 3T3 and HeLa, respectively), and others have shown that repression of translational initiation through the iron response element (a textbook example of mammalian translational repression) does not impart detectable destabilization of either its endogenous host mRNAs (Coccia et al., 1992; Melefors et al., 1993; Kim et al., 1996) or a reporter transcript (Hentze et al., 1987). Thus, when considered together, the available evidence strongly supports a model in which miRNAs actively recruit the deadenylation machinery, and the ensuing deadenylation, decapping, and decay comprises the major mode of miRNA-mediated repression of endogenous targets in mammalian cells.

Some translational repression accompanies mRNA destabilization as a minor component of endogenous target repression in mammalian cells. Like mRNA destabilization, this translational repression also appears to depend on recruitment of CCR4-NOT, but three observations indicate that this repression is not simply a consequence of shortened poly(A)-tails. First, mRNAs without poly(A) tails can be translationally repressed (Wu et al., 2006; Eulalio et al., 2008; Eulalio et al., 2009; Braun et al., 2011; Chekulaeva et al., 2011; Zekri et al., 2013). Second, mutant complexes lacking deadenylase activity can nonetheless promote translational repression (Cooke et al., 2010). Third, tail length and TE are not correlated in most mammalian settings (Subtelny et al., 2014). Thus, the two modes of miRNA-mediated repression seem to represent two independent ramifications of recruiting the deadenylation complexes.

Reconciling results with single-gene studies of mRNA and protein changes

The conclusion that mRNA destabilization is the major mode of miRNA-mediated repression agrees with many previous observations monitoring protein and mRNA changes of single target genes after perturbing a miRNA. Among the >30,000 research studies of mammalian miRNAs, there are also counter examples, in which single-gene measurements seem to suggest a greater role for translational repression (Poy et al., 2004; O'Donnell et al., 2005; Zhao et al., 2005; Chen et al., 2006). An advantage of our approach is that we simultaneously examine thousands of genes, comparing the changes of both mRNA level and TE for hundreds of genes that have at least one miRNA site to those of hundreds of genes that lack a site and thus serve as internal controls. The aggregate result of this global approach should reflect the overall contributions of mRNA destabilization and translational repression, whereas a single-gene study might choose a non-representative example and reach a conclusion that does not apply more generally to the targets of the miRNA.

This raises the question as to what might explain a single-gene result in which a miRNA-dependent change is observed in protein (i.e., with an immunoblot) but not mRNA (e.g., with quantitative reverse-transcription–PCR), which would appear as an outlier in our analyses. Might such outliers represent targets that are repressed at the level of translation without being destabilized? Although this possibility cannot be excluded, changes observed among our control genes that lack miRNA sites raise doubts about its validity. In most experiments (the possible exception being U2OS cells transfected with miR-155), a similar number of these control genes also change at the level of translation without being destabilized (Figure 1D). The observation that this behavior usually does not depend on the presence of a site to the miRNA suggests that either indirect effects of the miRNA or experimental variability explain the presence of most outliers that appear to be changing only at the level of translation.

Other single-gene examples for which translational repression is reported to be the major mode of miRNA-mediated regulation examine reporter mRNAs rather than endogenous mRNAs (Doench and Sharp, 2004; Kiriakidou et al., 2004; Nelson et al., 2004; Yekta et al., 2004; Pillai et al., 2005). Interestingly, the fractional component of regulation attributable to translational repression generally seems to be higher for reporters than for endogenous genes. We have begun experiments that aim to understand this difference between reporter and endogenous genes. Once this difference is understood, reporters could be developed that better recapitulate the regulation of endogenous genes, which would provide more relevant tools for studying the mechanism and dynamics of miRNA-mediated repression.

RNA Isolation

For RNA-seq, total RNA was extracted from B cells and U2OS cells using TRI reagent. Using TRI reagent, cytoplasmic RNA was extracted from cytoplasmic fractions of U2OS cells that were separated from nuclear fractions by differential centrifugation. Briefly, whole-cell lysate prepared as described (Guo et al., 2010) was centrifuged at 1,300g for 10 min, and the resulting supernatant was collected as the cytoplasmic fraction while the pellet obtained was collected as the nuclear fraction. To prepare rRNA/tRNA-depleted U2OS total RNA, total RNA was first treated with the Ribo-Zero rRNA removal kit (Epicentre BioTechnologies) according to manufacturer’s instructions. The resulting rRNA-depleted RNA sample was then spin-filtered using Ultra-4 centrifugal filters with Ultracel-100 membranes (Amicon) by centrifuging at 5,000g for 10 min at 4°C. The filtrate was enriched in tRNAs and was discarded, and the retentate was collected as the rRNA/tRNA-depleted RNA sample. RNA for all other RNA-seq samples was prepared by extracting RNA from ribosome profiling lysates with TRI reagent as described (Subtelny et al., 2014). Except in the case of the tRNA/rRNA-depleted U2OS RNA sample, the extracted RNA was poly(A)-selected as described (Subtelny et al., 2014).

Ribosome Footprint Profiling and RNA-seq

For B cell and U2OS samples, ribosome profiling and RNA-seq were performed essentially as described (Guo et al., 2010), with the only difference being how the RNA was isolated or enriched the cases of U2OS cytoplasmically enriched RNA and tRNA/rRNA-depleted total RNA. All other samples were prepared as described (Subtelny et al., 2014). Detailed protocols are available at http://bartellab.wi.mit.edu/protocols.html. Reference transcript annotations were downloaded (in refFlat format) from the UCSC Genome browser, and for each gene, the longest transcript was chosen as a representative transcript model. RPF and RNA-seq reads were mapped to ORFs as described, which excluded the first 50 nt of each ORF so as to eliminate signal from ribosomes that initiated after adding cycloheximide (Subtelny et al., 2014).

Accession numbers

The NCBI GEO accession number for all microarray and sRNA-seq data and most ribosome profiling and RNA-seq data is GSE61073. The accession number for HeLa and miR-223 neutrophil data analyzed in this study is GSE22004. The accession number for the U2OS ribosome profiling data and RNA-seq data from poly(A)-selected total RNA and tRNA/rRNA-depleted total RNA is GSE51584. The accession numbers for HEK293T mock-treated RNA-seq and ribosome profiling data are GSM1276541 and GSM1276542, respectively. The accession numbers for the uninduced miR-155 actively dividing 3T3 RNA-seq and ribosome profiling data are GSM1276543 and GSM1276544, respectively.